I came across the profile of a "Quantitative Marketing Analyst" on Linkedin. What's that?

Maybe this: http://mashable.com/2012/07/17/marketer ... cientists/

QuoteOriginally posted by: edouardI came across the profile of a "Quantitative Marketing Analyst" on Linkedin. What's that?Maybe border line or supporting business development dept? I know an ex-management consultant (physicist) hired in an equivalently named position, and what he does is basically market research, supported by some statistics and consumed by a bunch of managers for business development purposes.

The average guy who runs a discount carpet warehouse is smart enough to understand cost per click. But he is busy sorting out conflicts among the salary weirdos and scheming bitches who work for him. So he needs someone who also understands cost per click, to whom he can delegate managing the web-marketing presence. Only he cannot afford to pay too much, and he needs indians, not chiefs. So this would tend to be someone who is generally stupid, but with a narrow talent in basic arithmetic.

Thank you for your comments. The LI profile I saw had some taste of quantitative finance.

A large % of marketing people call what they do "quantitative" since they do lots of stats and now big data is a thing, that has expanded into trawling through databases.

QuoteOriginally posted by: DominicConnorA large % of marketing people call what they do "quantitative" since they do lots of stats and now big data is a thing, that has expanded into trawling through databases.Marketing people do not do Big Data, technology people do. And these are good technology people. Your resentment for Big Data seems related to the fact that it is based on Java than your beloved C++. And even though vendors link Big Data with Data Science, hardly any practitioner of Big Data calls himself a "Quantitative" guy who is considered one of narrower skills and hence less important or skilled.

Marketing people do not do Big Data, technology people do.That's not true, it barely even makes it to being false, certainly it is naive.Some marketing people do Big Data a lot, they don't necessarily do it well but many of them are mad keen on the idea, many, perhaps even most of the commercial Big Data projects out there came from people in marketing departments. I've been talking to those who hire Big Data analysts and they despair that so few people can both write the query and know what it means.And these are good technology people. Some are, some aren't, some marketing people could get many of the jobs advertised here, some are fit only to work at Deutsche Bank.Your resentment for Big Data seems related to the fact that it is based on Java than your beloved C++. Is someone else writing your posts for you ?You write English as if you can speak it properly, yet you clearly cannot comprehend the things I have written or the talks I have given saying that Big Data and related stuff is good for your CV and have even been accused of being part of the Big Data hype.As it happens a good chunk of the core tech is based on C++, so your English literacy issues are compounded by your technical illiteracy and worse you seem to think that the procedural programming involved is in any way relevant, it does not matter whether that is Java, C++, F# or VBA.And even though vendors link Big Data with Data Science, hardly any practitioner of Big Data calls himself a "Quantitative" guyOh dear, and I though you were merely illiterate...Try getting a grown up to explain my post to you and introduce you to LinkedIn.Even the simplest query of Linkedin such as searching for quantitative marketing will give hundreds of thousands of people who try (and sometimes succeed) to use quantitative methods for marketing. Many of them are trying to use Big Data methods, some will succeed because you will find (if you weren't both lazy and challenged by simple English) that many people in marketing have advanced degrees in things like statistics and yes before you ask some can do C++, though since you have problems understanding me I will repeat that this is largely irrelevant. Yes of course some marketing people are as technically illiterate as you, "marketing" covers a very wide range of activities, earlier this week three very pretty marketeers were engaged to form a rota such that at any point during the conference I was never on my own, was quite surreal since they understood at best 1 word in 4 in the conversations they were supposed to be moderating.

QuoteOriginally posted by: DominicConnorMarketing people do not do Big Data, technology people do.That's not true, it barely even makes it to being false, certainly it is naive.Some marketing people do Big Data a lot, they don't necessarily do it well but many of them are mad keen on the idea, many, perhaps even most of the commercial Big Data projects out there came from people in marketing departments. I've been talking to those who hire Big Data analysts and they despair that so few people can both write the query and know what it means.And these are good technology people. Some are, some aren't, some marketing people could get many of the jobs advertised here, some are fit only to work at Deutsche Bank.Your resentment for Big Data seems related to the fact that it is based on Java than your beloved C++. Is someone else writing your posts for you ?You write English as if you can speak it properly, yet you clearly cannot comprehend the things I have written or the talks I have given saying that Big Data and related stuff is good for your CV and have even been accused of being part of the Big Data hype.As it happens a good chunk of the core tech is based on C++, so your English literacy issues are compounded by your technical illiteracy and worse you seem to think that the procedural programming involved is in any way relevant, it does not matter whether that is Java, C++, F# or VBA.And even though vendors link Big Data with Data Science, hardly any practitioner of Big Data calls himself a "Quantitative" guyOh dear, and I though you were merely illiterate...Try getting a grown up to explain my post to you and introduce you to LinkedIn.Even the simplest query of Linkedin such as searching for quantitative marketing will give hundreds of thousands of people who try (and sometimes succeed) to use quantitative methods for marketing. Many of them are trying to use Big Data methods, some will succeed because you will find (if you weren't both lazy and challenged by simple English) that many people in marketing have advanced degrees in things like statistics and yes before you ask some can do C++, though since you have problems understanding me I will repeat that this is largely irrelevant. Yes of course some marketing people are as technically illiterate as you, "marketing" covers a very wide range of activities, earlier this week three very pretty marketeers were engaged to form a rota such that at any point during the conference I was never on my own, was quite surreal since they understood at best 1 word in 4 in the conversations they were supposed to be moderating.Been through many of these and they get boorish. I am going to pass on this provocation.

My 2 cents (based purely on gut feeling and 1 book I read,so take with a spoon of salt) is that languages such as Java (and C# maybe) are more suitable than C++ for these kinds of applications in much the same way that C++ for database applications a PITA is. IMO C# is much easier. If I were the decision maker I would ensure that all alternatives be investigated. I wonder what the status in C++ is of MapReduce// I hope this does not start the olde language wars again.// What a pity it can't be discussed like rational and polite beings.

QuoteOriginally posted by: CuchulainnMy 2 cents (based purely on gut feeling and 1 book I read,so take with a spoon of salt) is that languages such as Java (and C# maybe) are more suitable than C++ for these kinds of applications in much the same way that C++ for database applications a PITA is. IMO C# is much easier. If I were the decision maker I would ensure that all alternatives be investigated. I wonder what the status in C++ is of MapReduce// I hope this does not start the olde language wars again.// What a pity it can't be discussed like rational and polite beings.There is something from LexisNexis on C++ for High Performance Computing which is supposed to compete with Hadoop. I was invited once at NIST (National Institute of Science and Technology) for a one day workshop but had to skip it. Couple of links (I have not read them) - 1. Hadoop Killer2. HPCC OverviewHortonworks has released Hadoop for .NET. It is a very recent phenomenon and in my opinion a very welcome one from the point of view of expanding the native language support for Map Reduce.There is always Hadoop Streaming which you can use with any language but it does generate a seperate sub-process of each of the Map and Reduce processes. This is fine except one needs to remember that the HotSpot JVM was implemented to use the fork() instead of vfork() leading to twice the memory being required as the parent process page table is cloned unnecessarily. Managing the vm_overcommit_memory=1 and vm_overcommit_ratio is required to ensure there is no running out of memory even when there is plenty. I have seen Hadoop Streaming being used when Hadoop is used to parallelize I/O operations whereas CUDA is used to improve CPU performance.

capafan2,I saw this; is it feasible ?QuoteHere are some examples of awesome things you should be able to do with a Big Data exploratory analytics database. 1. Build the ARCA book for one day of all exchange-traded US equities (186 million quotes) in 80 seconds on a 32-instance commodity hardware cluster. Run it in about half the time on a cluster twice as large.2. Run a Principle Components Analysis on a 50M x 50M sparse matrix in minutes.3. Select data sets (based on complex criteria) in constant time?irrespective of how big your dataset gets.

QuoteOriginally posted by: Cuchulainncapafan2,I saw this; is it feasible ?QuoteHere are some examples of awesome things you should be able to do with a Big Data exploratory analytics database. 1. Build the ARCA book for one day of all exchange-traded US equities (186 million quotes) in 80 seconds on a 32-instance commodity hardware cluster. Run it in about half the time on a cluster twice as large.2. Run a Principle Components Analysis on a 50M x 50M sparse matrix in minutes.3. Select data sets (based on complex criteria) in constant time?irrespective of how big your dataset gets.Some of the claims you see attributed to MapReduce are tall. However, if you design well and take into account all the special characteristics in your data and apply appropriate partitioning opportunities provided by the data good performance can be achieved (Ex. Partitioned Datasets can be executed on seperate MapReduce clusters in parallel. This is even more real in the Cloud where you can provision as many resources in seperate clusters if you like for short duration and get data from a common and cheap persistent source , ex. Amazon S3).The other main criteria is how linear and partionable is the process (PCA is probably a good example of such a process). See the text book - Mining Massive Datasets from Stanford. Interestingly it looks at several examples from Quantitative Marketing (Ex. Collaborative Filtering). Chapter two talks about MapReduce for sparse datasets and it is quick read.Also the more you do in the Map processes of the MapReduce process the more linearly scalable you process with respect to nodes. As soon as you need a Reduce phase, it becomes dependent on the amount of data required to be moved to the reduce side, and there is a natural bottleneck in the Shuffle and Sort process which precedes Reduce. More design effort is needed in this case.Several processes which are naturally Map+Reduce can be converted into Map Only. Let's say you have two very large datasets(0.5 Billion each) and you wish to join on a simple criteria. The way it is made Map Only is by chaining two Map Processes-1. Sort both Dataset based on Join Criteria (ex. A.x and B.y)2. Next a Map Only Process (and hence very scalable) will run through B.y and create index of starting key and ending key (B.y values) of each block storing sorted B3. Next a Map Only Process will run through A.x and as each value is encountered make a decision based on the index developed in (2) to load the appropriate block in memory (usually about 64 MB or 128 MB large) and perform the join then and there.Basically a Sort-Merge-Join operation in practice. In one case I got through about 1.2 Billion rows on both sides of a join in under 15 minutes (could be 10, don't remember well) on a 30 node cluster (including sorting)Selecting based on complex criteria is possible as long as the number of joins is not too many or even more than 3. Complex where clause on a single table will scale easily but if extra data is needed to make the decision the main criteria is can this entire extra data be loaded once in each Map Process without running out of memory (called Distributed Cache in Hadoop). I have used Memory Mapped Files here to avoid Garbage Collection problems.And there are other frameworks like Apache Hama (BSP model) which is ideal for Machine Learning iterative algos, Spark - (http://spark.incubator.apache.org/ ) is a good framework for this as well. Both reduce I/O between iterations. Really depends on your problem. Most solutions in this domain are very custom as in my opinion if you do not take advantage of the specific characteristics in your data, things take orders of magnitude more time.

And what does a Derivative Commercial Analyst?

A skill or talent that is both hard to find as well as hard to measure is whether you can ask the right questions, it is not enough to master the tech and the maths.At one conference recently the speaker showed how he could predict what a white male between the ages of 31 and 35 with a graduate degree was most likely to buy when he visited a supermarket.Bread ,milk and eggs....Because of course, those are the most common purchases for everyone.That's easy to laugh at because they familiar patterns, but when the outputs are things like which part of a websitre are accessed or some abtruse pattern in electricity usage, it is a hole that's easy ot fall into.An extra problem is that prediction is rarely what we want in this context, it is changing behaviour, what would make him buy some premium coffee ? We want that because the margins on Bread, etc are wafer thin and often negative, someone who just buys what they need is not an ideal customer.I suspect (but don't know) that a Derivative Commercial Analyst is someone who works out how to pitch and price hedges,caps and swaps to customers of the energy firm.